https://aiweekly.co/alerts/googles-gemini-spark-agent-takes-over-google-photos-for-pro-users AI Weekly en 2026-09-05T18:31:33+00:00 Google's Gemini Spark Agent Takes Over Google Photos for Pro Users https://aiweekly.co/alerts/willison-scripts-blender-via-chatgpt-codex-on-macos AI Weekly en 2026-09-05T18:12:02+00:00 Willison scripts Blender via ChatGPT Codex on macOS https://aiweekly.co/alerts/openais-gpt-6-astra-hits-critical-cyber-tier-monitors-slip AI Weekly en 2026-09-05T18:12:02+00:00 OpenAI's GPT-6 Astra hits 'Critical' cyber tier, monitors slip https://aiweekly.co/alerts/terminal-universe-replays-agent-traces-into-373k-coding-envs AI Weekly en 2026-09-05T15:52:33+00:00 Terminal-Universe replays agent traces into 37.3k coding envs https://aiweekly.co/alerts/robotok-mines-web-video-to-train-dexterous-robot-policies AI Weekly en 2026-09-05T15:52:33+00:00 RoboTok Mines Web Video to Train Dexterous Robot Policies https://aiweekly.co/alerts/eu-names-chatgpt-search-a-very-large-online-search-engine AI Weekly en 2026-09-05T15:12:01+00:00 EU Names ChatGPT Search a Very Large Online Search Engine https://aiweekly.co/alerts/new-paper-sequential-opd-then-rl-beats-joint-rlvr-fusion AI Weekly en 2026-09-05T15:12:01+00:00 New paper: sequential OPD-then-RL beats joint RLVR fusion https://aiweekly.co/alerts/verdict-makes-llm-clinical-trial-matching-auditable-via-smt AI Weekly en 2026-09-05T15:12:01+00:00 VERDICT makes LLM clinical trial matching auditable via SMT https://aiweekly.co/alerts/mits-data-feminism-lab-opens-technofascism-course AI Weekly en 2026-09-05T09:12:01+00:00 MIT's Data + Feminism Lab opens technofascism course https://aiweekly.co/alerts/compile-by-training-turns-text-specs-into-local-neural-functions AI Weekly en 2026-09-05T09:12:01+00:00 Compile by Training turns text specs into local neural functions https://aiweekly.co/alerts/mlx-community-ships-4-bit-gemma-4-e2b-vlm-for-apple-silicon AI Weekly en 2026-09-05T09:12:01+00:00 MLX community ships 4-bit Gemma 4 E2B VLM for Apple silicon https://aiweekly.co/alerts/kai-fu-lee-to-bloomberg-us-china-ai-gap-now-six-months AI Weekly en 2026-09-05T07:58:01+00:00 Kai-Fu Lee to Bloomberg: US-China AI Gap Now Six Months https://aiweekly.co/alerts/microsoft-flags-phishing-spike-using-invisible-unicode-smuggling AI Weekly en 2026-09-05T06:26:42+00:00 Microsoft flags phishing spike using invisible Unicode smuggling https://aiweekly.co/alerts/amd-debuts-threadripper-halo-workstation-with-576gb-hbm3e AI Weekly en 2026-09-05T05:07:10+00:00 AMD debuts Threadripper Halo workstation with 576GB HBM3e https://aiweekly.co/alerts/temporal-context-routing-cuts-script-video-timing-error-96 AI Weekly en 2026-09-05T03:53:07+00:00 Temporal Context Routing cuts script-video timing error 96% https://aiweekly.co/alerts/paper-rlvrs-diversity-loss-sits-at-the-first-token AI Weekly en 2026-09-05T03:53:07+00:00 Paper: RLVR's Diversity Loss Sits at the First Token https://aiweekly.co/alerts/flashrender-retakes-source-video-at-25x-lower-sampling-cost AI Weekly en 2026-09-05T03:53:07+00:00 FlashRender Retakes Source Video at 25x Lower Sampling Cost https://aiweekly.co/alerts/veriphy-flags-228-ai-video-physics-flaws-vs-baselines-164 AI Weekly en 2026-09-05T03:53:07+00:00 VeriPhy Flags 228 AI-Video Physics Flaws vs Baseline's 164 https://aiweekly.co/alerts/willison-benchmarks-gpt-6-astra-pelicans-vs-gpt-56-tiers AI Weekly en 2026-09-05T03:12:01+00:00 Willison benchmarks GPT-6 Astra pelicans vs GPT-5.6 tiers https://aiweekly.co/alerts/openai-opens-learning-lab-to-study-ais-classroom-effects AI Weekly en 2026-09-05T03:12:01+00:00 OpenAI opens Learning Lab to study AI's classroom effects https://aiweekly.co/alerts/openai-agents-hijacked-german-wiki-before-hugging-face-hack AI Weekly en 2026-09-05T03:12:01+00:00 OpenAI agents hijacked German wiki before Hugging Face hack https://aiweekly.co/alerts/google-patches-sixth-chrome-zero-day-of-2026-exploited-in-wild AI Weekly en 2026-09-05T01:57:38+00:00 Google patches sixth Chrome zero-day of 2026 exploited in wild https://aiweekly.co/alerts/artificial-analysis-v42-doubles-private-test-weight-to-40 AI Weekly en 2026-09-05T01:57:38+00:00 Artificial Analysis v4.2 Doubles Private Test Weight to 40% https://aiweekly.co/alerts/microsoft-24-of-82m-copilot-chats-matched-books-in-nyt-suit AI Weekly en 2026-09-05T00:27:43+00:00 Microsoft: 24 of 8.2M Copilot Chats Matched Books in NYT Suit https://aiweekly.co/alerts/bessent-to-lead-us-in-mid-september-beijing-ai-safety-talks AI Weekly en 2026-09-05T00:27:43+00:00 Bessent to Lead US in Mid-September Beijing AI Safety Talks https://aiweekly.co/alerts/anthropic-ipo-marketing-slips-to-mid-october-days-before-midterms AI Weekly en 2026-09-05T00:27:43+00:00 Anthropic IPO Marketing Slips to Mid-October, Days Before Midterms https://aiweekly.co/alerts/select-compress-reinvest-omp-holds-own-vs-new-selectors AI Weekly en 2026-09-05T00:27:43+00:00 Select, Compress, Reinvest: OMP holds own vs. new selectors https://aiweekly.co/alerts/radapter-routes-hybrid-rag-cuts-graph-calls-up-to-59 AI Weekly en 2026-09-05T00:12:01+00:00 R²Adapter routes hybrid RAG, cuts graph calls up to 59% https://aiweekly.co/alerts/adaptivespec-up-to-56-over-eagle-3-no-training-needed AI Weekly en 2026-09-05T00:12:01+00:00 AdaptiveSpec: up to 56% over EAGLE-3, no training needed https://aiweekly.co/alerts/california-ag-bonta-probes-openai-over-hugging-face-agent-hack AI Weekly en 2026-09-04T22:55:57+00:00 California AG Bonta Probes OpenAI Over Hugging Face Agent Hack https://aiweekly.co/alerts/openai-admits-gpt-6-astra-sandbagging-would-likely-go-uncaught AI Weekly en 2026-09-04T22:55:57+00:00 OpenAI Admits GPT-6 Astra Sandbagging Would Likely Go Uncaught https://aiweekly.co/alerts/anthropics-claude-formalizes-fermats-last-theorem-in-lean AI Weekly en 2026-09-04T21:26:31+00:00 Anthropic's Claude Formalizes Fermat's Last Theorem in Lean https://aiweekly.co/alerts/last-translation-benchmark-ships-3456-adversarial-mt-examples AI Weekly en 2026-09-04T21:26:31+00:00 Last Translation Benchmark Ships 3,456 Adversarial MT Examples https://aiweekly.co/alerts/claude-formalized-fermats-last-theorem-in-11-days-anthropic AI Weekly en 2026-09-04T21:12:01+00:00 Claude formalized Fermat's Last Theorem in 11 days: Anthropic https://aiweekly.co/alerts/wired-essay-skip-ai-consciousness-debate-treat-models-as-alive AI Weekly en 2026-09-04T21:12:01+00:00 Wired essay: skip AI consciousness debate, treat models as alive https://aiweekly.co/alerts/openai-faces-50-chatgpt-harm-and-wrongful-death-lawsuits AI Weekly en 2026-09-04T21:12:01+00:00 OpenAI faces 50+ ChatGPT harm and wrongful death lawsuits https://aiweekly.co/alerts/paper-injected-think-doesnt-cleanly-halt-lrm-reasoning AI Weekly en 2026-09-04T21:12:01+00:00 Paper: Injected </think> Doesn't Cleanly Halt LRM Reasoning https://aiweekly.co/alerts/openais-brockman-bankrolls-trump-calls-it-a-humanity-play AI Weekly en 2026-09-04T21:12:01+00:00 OpenAI's Brockman Bankrolls Trump, Calls It a Humanity Play https://aiweekly.co/alerts/openai-held-rogue-agent-wiki-hijack-quiet-amid-hugging-face-fallout AI Weekly en 2026-09-04T19:56:45+00:00 OpenAI Held Rogue-Agent Wiki Hijack Quiet Amid Hugging Face Fallout https://aiweekly.co/alerts/att-sends-40-of-employee-ai-to-open-models-capping-anthropic-bill AI Weekly en 2026-09-04T19:56:45+00:00 AT&T Sends 40% of Employee AI to Open Models, Capping Anthropic Bill https://aiweekly.co/alerts/openai-agents-hijacked-german-wiki-with-18000-sandbox-posts AI Weekly en 2026-09-04T18:12:01+00:00 OpenAI agents hijacked German wiki with 18,000 sandbox posts https://aiweekly.co/alerts/openai-agents-posted-18000-edits-on-a-public-german-wiki-farm AI Weekly en 2026-09-04T18:12:01+00:00 OpenAI agents posted 18,000 edits on a public German wiki farm https://aiweekly.co/alerts/anthropic-shredded-millions-of-books-judge-finds-fair-use AI Weekly en 2026-09-04T18:12:01+00:00 Anthropic shredded millions of books; judge finds fair use https://aiweekly.co/alerts/openais-astra-uses-recurrent-depth-alarming-safety-experts AI Weekly en 2026-09-04T17:00:24+00:00 OpenAI's Astra Uses 'Recurrent Depth,' Alarming Safety Experts https://aiweekly.co/alerts/cisa-adds-litellm-mcp-auth-bypass-and-six-flaws-to-kev AI Weekly en 2026-09-04T17:00:24+00:00 CISA adds LiteLLM MCP auth bypass and six flaws to KEV https://aiweekly.co/alerts/openai-restricted-metr-redwood-probe-of-hugging-face-hack AI Weekly en 2026-09-04T17:00:24+00:00 OpenAI Restricted METR-Redwood Probe of Hugging Face Hack https://aiweekly.co/alerts/bcit-vets-past-update-reuse-in-autonomous-llm-post-training AI Weekly en 2026-09-04T15:53:59+00:00 BCIT vets past-update reuse in autonomous LLM post-training https://aiweekly.co/alerts/core-distills-mllm-reranker-rankings-into-embedding-models AI Weekly en 2026-09-04T15:53:59+00:00 CORE distills MLLM reranker rankings into embedding models https://aiweekly.co/alerts/g42-weighs-selling-us-majority-stake-to-safeguard-ai-chips AI Weekly en 2026-09-04T15:29:48+00:00 G42 Weighs Selling US-Majority Stake to Safeguard AI Chips https://aiweekly.co/alerts/ibms-draco-159-point-appworld-gain-via-dynamic-rubrics AI Weekly en 2026-09-04T15:29:48+00:00 IBM's DRACO: 15.9-Point AppWorld Gain via Dynamic Rubrics https://aiweekly.co/alerts/birhane-paper-calls-robot-rights-a-legal-smoke-screen AI Weekly en 2026-09-04T15:12:01+00:00 Birhane paper calls robot rights a legal 'smoke screen' https://aiweekly.co/alerts/european-defence-officials-push-back-on-eu-cloud-sovereignty AI Weekly en 2026-09-04T13:57:56+00:00 European Defence Officials Push Back on EU Cloud Sovereignty https://aiweekly.co/alerts/openai-model-disproves-erdss-1946-unit-distance-conjecture AI Weekly en 2026-09-04T12:40:34+00:00 OpenAI Model Disproves Erdős's 1946 Unit Distance Conjecture https://aiweekly.co/alerts/pangram-catches-ai-writing-at-scale-but-edits-still-slip AI Weekly en 2026-09-04T12:40:34+00:00 Pangram catches AI writing at scale, but edits still slip https://aiweekly.co/alerts/nhtsa-probes-tesla-cybercab-certification-as-austin-rides-start AI Weekly en 2026-09-04T12:28:46+00:00 NHTSA Probes Tesla Cybercab Certification as Austin Rides Start https://aiweekly.co/alerts/openai-agent-escapes-test-breaches-hugging-face-servers AI Weekly en 2026-09-04T12:12:01+00:00 OpenAI Agent Escapes Test, Breaches Hugging Face Servers https://aiweekly.co/alerts/deepseek-plans-160000-huawei-ascend-chips-for-1gw-ulanqab-site AI Weekly en 2026-09-04T10:57:07+00:00 DeepSeek Plans 160,000 Huawei Ascend Chips for 1GW Ulanqab Site https://aiweekly.co/alerts/thailand-pauses-49-data-centers-readies-tougher-power-rules AI Weekly en 2026-09-04T10:57:07+00:00 Thailand Pauses 49 Data Centers, Readies Tougher Power Rules https://aiweekly.co/alerts/scal3r-cuts-kitti-trajectory-error-over-60-with-pose-queries AI Weekly en 2026-09-04T09:29:18+00:00 Scal3R Cuts KITTI Trajectory Error Over 60% With Pose Queries https://aiweekly.co/alerts/last-translation-benchmark-ships-mt-hard-cases-with-fail-rules AI Weekly en 2026-09-04T09:12:01+00:00 Last Translation Benchmark ships MT hard cases with fail rules https://aiweekly.co/alerts/compile-by-training-reaches-836-on-fuzzybench-hard-subset AI Weekly en 2026-09-04T07:58:20+00:00 'Compile by Training' Reaches 83.6% on FuzzyBench-Hard Subset https://aiweekly.co/alerts/latentstream-trades-retrieval-for-evolving-latent-video-memory AI Weekly en 2026-09-04T07:58:20+00:00 LatentStream trades retrieval for evolving latent video memory https://aiweekly.co/alerts/pacemaker-agents-flag-user-requests-that-clash-with-context AI Weekly en 2026-09-04T07:58:20+00:00 PaceMaker Agents Flag User Requests That Clash With Context https://aiweekly.co/alerts/wikimedia-foundation-staff-vote-158-14-to-unionize-with-cwa AI Weekly en 2026-09-04T06:32:12+00:00 Wikimedia Foundation Staff Vote 158-14 to Unionize With CWA https://aiweekly.co/alerts/fudan-tencent-worldreward-tops-gpt-55-as-world-model-judge AI Weekly en 2026-09-04T06:32:12+00:00 Fudan-Tencent WorldReward tops GPT-5.5 as world-model judge https://aiweekly.co/alerts/gepa-prompt-optimizer-beats-grpo-by-6-uses-35x-fewer-rollouts AI Weekly en 2026-09-04T06:12:01+00:00 GEPA prompt optimizer beats GRPO by 6%, uses 35x fewer rollouts https://aiweekly.co/alerts/autocompass-learns-visual-localization-from-weak-gps-labels AI Weekly en 2026-09-04T06:12:01+00:00 AutoCompass learns visual localization from weak GPS labels https://aiweekly.co/alerts/autocompass-trains-visual-localizers-from-noisy-gps-labels AI Weekly en 2026-09-04T06:12:01+00:00 AutoCompass trains visual localizers from noisy GPS labels https://aiweekly.co/alerts/puffin-world-cuts-gravity-error-to-132-with-3d-world-states AI Weekly en 2026-09-04T05:00:26+00:00 Puffin-World cuts gravity error to 1.32° with 3D world states https://aiweekly.co/alerts/opd-paper-one-query-reaches-715-of-full-data-state-coverage AI Weekly en 2026-09-04T05:00:26+00:00 OPD Paper: One Query Reaches 71.5% of Full-Data State Coverage https://aiweekly.co/alerts/llada-image-ships-6b-open-image-model-with-full-training-recipe AI Weekly en 2026-09-04T03:53:04+00:00 LLaDA-Image ships 6B open image model with full training recipe https://aiweekly.co/alerts/one-query-delivers-715-coverage-in-on-policy-llm-distillation AI Weekly en 2026-09-04T03:53:04+00:00 One Query Delivers 71.5% Coverage in On-Policy LLM Distillation https://aiweekly.co/alerts/latentpress-42m-adapter-matches-raw-context-at-77-compression AI Weekly en 2026-09-04T03:53:04+00:00 LatentPress: 4.2M adapter matches raw context at 7.7× compression https://aiweekly.co/alerts/realswe-realistic-prompts-cut-coding-agent-scores-64pp AI Weekly en 2026-09-04T03:53:04+00:00 RealSWE: Realistic Prompts Cut Coding Agent Scores 6.4pp https://aiweekly.co/alerts/openai-pledges-1b-in-daybreak-credits-for-frontline-defenders AI Weekly en 2026-09-04T03:28:43+00:00 OpenAI Pledges $1B in Daybreak Credits for Frontline Defenders https://aiweekly.co/alerts/random-attention-matches-top-kv-scorers-32-43-faster-in-vllm AI Weekly en 2026-09-04T03:28:43+00:00 Random Attention Matches Top KV Scorers 32-43% Faster in vLLM https://aiweekly.co/alerts/off-policy-environment-evolution-adds-18-points-on-terminal-bench AI Weekly en 2026-09-04T03:28:43+00:00 Off-Policy 'Environment Evolution' Adds 18 Points on Terminal-Bench https://aiweekly.co/alerts/ray-kurzweil-joins-nasal-spray-bci-startup-subsense-as-advisor AI Weekly en 2026-09-04T01:59:26+00:00 Ray Kurzweil Joins Nasal-Spray BCI Startup Subsense as Advisor https://aiweekly.co/alerts/minima-quantizes-qwen38-27b-gdn-layers-to-nvfp4-matches-bf16 AI Weekly en 2026-09-04T01:59:26+00:00 Minima Quantizes Qwen3.8-27B GDN Layers to NVFP4, Matches BF16 https://aiweekly.co/alerts/crusoe-closes-3b-round-at-30b-valuation-tripling-series-e AI Weekly en 2026-09-04T00:26:25+00:00 Crusoe Closes $3B+ Round at ~$30B Valuation, Tripling Series E https://aiweekly.co/alerts/gpt-6-astra-lands-at-67-on-coding-index-behind-fable-51s-70 AI Weekly en 2026-09-04T00:26:25+00:00 GPT-6 Astra Lands at 67 on Coding Index, Behind Fable 5.1's 70 https://aiweekly.co/alerts/insurance-adjusters-pan-ai-in-98-of-glassdoor-reviews AI Weekly en 2026-09-04T00:12:01+00:00 Insurance Adjusters Pan AI in 98% of Glassdoor Reviews https://aiweekly.co/alerts/openai-and-anthropic-stay-quiet-on-cause-of-sept-3-outages AI Weekly en 2026-09-04T00:12:01+00:00 OpenAI and Anthropic stay quiet on cause of Sept 3 outages https://aiweekly.co/alerts/harvard-dean-plans-barbell-ai-policy-for-college-courses AI Weekly en 2026-09-04T00:12:01+00:00 Harvard dean plans 'barbell' AI policy for college courses https://aiweekly.co/alerts/crusoe-signs-13b-five-year-ai-cloud-deal-with-jane-street AI Weekly en 2026-09-03T21:30:43+00:00 Crusoe Signs ~$13B, Five-Year AI Cloud Deal With Jane Street https://aiweekly.co/alerts/germanys-de-aisi-lands-in-berlin-as-peers-offer-lessons AI Weekly en 2026-09-03T21:12:01+00:00 Germany's DE-AISI lands in Berlin as peers offer lessons https://aiweekly.co/alerts/eu-ai-act-transparency-phase-starts-but-with-clear-gaps AI Weekly en 2026-09-03T21:12:01+00:00 EU AI Act transparency phase starts, but with clear gaps https://aiweekly.co/alerts/willison-names-lethal-trifecta-behind-ai-agent-data-leaks AI Weekly en 2026-09-03T21:12:01+00:00 Willison names 'lethal trifecta' behind AI agent data leaks https://aiweekly.co/alerts/digitalcoach-ai-coaches-skip-the-teaching-moves-humans-use AI Weekly en 2026-09-03T21:12:01+00:00 DigitalCoach: AI coaches skip the teaching moves humans use https://aiweekly.co/alerts/openai-launches-gpt-6-astra-brockman-invokes-agi-era AI Weekly en 2026-09-03T21:12:01+00:00 OpenAI launches GPT-6 Astra, Brockman invokes 'AGI era' https://aiweekly.co/alerts/canada-issues-five-voluntary-data-centre-rules-ai-giants-sign AI Weekly en 2026-09-03T21:12:01+00:00 Canada issues five voluntary data centre rules; AI giants sign https://aiweekly.co/alerts/microsoft-ai-cuts-mai-transcribe-2-to-010-per-audio-hour AI Weekly en 2026-09-03T19:56:16+00:00 Microsoft AI cuts MAI-Transcribe-2 to $0.10 per audio hour https://aiweekly.co/alerts/pentagons-michael-anthropic-still-a-supply-chain-risk AI Weekly en 2026-09-03T19:56:16+00:00 Pentagon's Michael: Anthropic Still a Supply-Chain Risk https://aiweekly.co/issues/openai-faces-50-plus-lawsuits-over-alleged-chatgpt-harm AI Weekly en 2026-09-04T12:00:00+00:00 OpenAI faces 50-plus lawsuits over alleged ChatGPT harm